Ore Grade Estimation with Modular Neural Network Systems – A Case Study

This paper introduces a neural network approach to the problem of ore grade estimation. The system under consideration consists of three neural network modules each responsible for a different area of the deposit, depending on the sampling density. Octant and quadrant search is used as a way of presenting input patterns to the modules. Both radial basis function networks and multi-layered perceptrons are used as the building blocks of these modules. An iron ore deposit provides the training and testing data for both the neural network system and kriging, and the results from the two approaches are compared.